Reviewers ask for analyses that don't fit. Study sections doubt the power analysis. Models fail to converge, and results contradict the literature. Each Case in the Clinic explains what the symptom means, what usually causes it, which checks to run, and how to explain the decision.
Choose the problem closest to yours to narrow the cases below.
Post hoc power · multiplicity · model choice
Browse 3 cases →Pilot data · ICCs · attrition
Browse 2 cases →Convergence · random effects · estimation
Browse 2 cases →Sign flips · suppression · interactions
Browse 2 cases →Missing data · clusters · measurement
Browse 1 cases →Reproducibility · software · workflow
Browse 1 cases →These educational cases use synthetic data, simulations, or specified numerical illustrations. They are not client case reports. Check each case’s assumptions before applying its approach to your study.
All 11 cases
Why "observed power" can't answer the reviewer's real question, and the three analyses that can.
Whether to correct depends on the claim you're making, not on how many p-values are in the paper.
Small pilots give effect sizes too noisy to plan on, and the pilots that look most promising are the most misleading.
Choose an analysis that respects the unit of assignment and dependence, then show what changes in the inference.
Show how the required number of clusters changes across plausible ICCs, and separate participant attrition from cluster loss.
Separate numerical trouble, boundary estimates, and unsupported random slopes before changing the model.
Check the analysis rows and the adjustment question before treating a sign reversal as evidence of a mistake.
Make the missing-data assumptions visible, choose a method that matches the study, and show sensitivity to departures from MAR.
Trace the first point of divergence in inputs, analysis rows, coding, and software before comparing final p-values.
Investigate perfect prediction before interpreting huge odds ratios or treating an optimizer’s last iterate as an estimate.
Translate the interaction into planned slopes and contrasts, with uncertainty on the scale your research question needs.
A quick fix can make a warning disappear or a comment go away while leaving the inference weaker than before. Each Case explains what the problem means for your conclusions, compares the defensible responses, and gives you wording you can adapt. Every number comes from an R script you can download and run.
Working on a dissertation? Dissertation Stats Helper's Analysis Clinic covers the problems doctoral students meet most. Preparing a manuscript? Reviewer Ready reviews the statistics before you submit.
Record your question, warning or reviewer comment, checks attempted, and next decision in the free troubleshooting worksheet.
Open the printable worksheet →Some problems turn on the details of your design, data, or the exact reviewer comment. A free 30-minute consult can identify the next defensible step and what it would take.